xindongzhang / mbmd

If you fail to install and run this tracker, please email me ([email protected])

Introduction

This repository includes tensorflow code of MBMD (MobileNet-based tracking by detection algorithm) for VOT2018 Long-Term Challenge.
The corresponding arxiv paper has been drafted on Arxiv.
Learning regression and verification networks for long-term visual tracking.
Yunhua Zhang, Dong Wang, Lijun Wang, Jinqing Qi, Huchuan Lu

Prerequisites

python 2.7
ubuntu 14.04
cuda-8.0
cudnn-6.0.21
Tensorflow-1.3-gpu
NVIDIA TITAN X GPU

Pretrained model

The bounding box regression's architecture is MobileNet, and the verifier's architecture is VGGM.
The pre-trained model can be downloaded at https://drive.google.com/open?id=1g3aMRi6CWK88FOEYoQjqs61fY6QvGW1Z.
Then you should copy the two files to the folder of our code.

Integrate into VOT-2018

The interface for integrating the tracker into the vot evaluation tool kit is implemented in the module python_long_MBMD.py. The script tracker_MBMD.m is needed to be copied to vot-tookit.

CPU manner

If you want to run this code on CPU, you need to just set os.environ ["CUDA_VISIBLE_DEVICES"]="" in the begin of python_long_MBMD.py